Template Bridge Plugin
Unified development workflow for Claude Code: Superpowers + Beads + Templates.
Bridges three powerful Claude Code plugins into a single disciplined workflow with on-demand access to 413+ specialist agent templates.
Prerequisites
| Plugin | Install | Purpose |
|---|---|---|
| Superpowers | /plugin install superpowers | TDD, brainstorming, debugging, code review |
| Beads | /plugin install beads | Task tracking across sessions |
Both are optional — the workflow skill adapts, but you get the full power with all three.
What's Inside
| Component | Type | Purpose |
|---|---|---|
unified-workflow | Skill | How to work: beads task → brainstorm → plan → TDD → review → finish |
template-catalog | Skill | 413+ specialist agents on-demand from claude-code-templates |
browse-templates | Command | /browse-templates — manual template search |
The Stack
┌─────────────────────────────────────────────────────┐
│ YOUR TASK │
├──────────┬──────────────────┬───────────────────────┤
│ Beads │ Superpowers │ Template Bridge │
│ WHAT │ HOW │ WHO │
│ │ │ │
│ bd create│ brainstorming │ template-catalog │
│ bd ready │ writing-plans │ 413+ specialist agents│
│ bd close │ TDD (red-green) │ on-demand install │
│ │ debugging │ │
│ │ code-review │ │
│ │ verification │ │
└──────────┴──────────────────┴───────────────────────┘
The Flow
Every task follows this sequence:
1. TASK bd create -t feature "..." or bd update <id> --status in_progress
│
2. BRAINSTORM superpowers:brainstorming → design before code
│
3. PLAN superpowers:writing-plans → 2-5 min tasks
│
4. ISOLATE superpowers:using-git-worktrees (non-trivial work)
│
5. IMPLEMENT For each task:
├─ RED: write ONE failing test
├─ VERIFY: confirm it fails for the right reason
├─ GREEN: minimal code to pass
├─ VERIFY: all tests pass
├─ REFACTOR: clean up
└─ COMMIT
│
6. REVIEW superpowers:requesting-code-review
│
7. VERIFY superpowers:verification-before-completion
│
8. FINISH superpowers:finishing-a-development-branch
│
9. CLOSE bd close <id> --reason "Done"
TDD: Red-Green-Refactor
No production code without a failing test first. Ever.
RED ──────► VERIFY RED ──────► GREEN ──────► VERIFY GREEN ──────► REFACTOR ──────► COMMIT
write test fails for minimal all tests clean up after each
right reason code pass green
Debugging
Always invoke superpowers:systematic-debugging BEFORE any fix:
- Root cause — read errors, reproduce, check recent changes
- Pattern analysis — find working examples, compare
- Hypothesis — single-variable test
- Fix — with failing test first
- Stop after 3 failed attempts → question architecture
Verification Gate
Before claiming anything done:
- Identify the proving command
- Run it fresh
- Read complete output + exit code
- Verify it supports the claim
- Only then claim done
"Evidence before claims, always."
External Templates
When a task needs a specialist not in current skills:
# Direct install (always fresh from repo)
npx claude-code-templates@latest --agent security/security-auditor --yes
# Interactive browser
npx claude-code-templates@latest
Categories (26)
| Category | Top Agents |
|---|---|
| ai-specialists | prompt-engineer, llm-architect, model-evaluator |
| api-graphql | api-architect, graphql-specialist |
| blockchain-web3 | smart-contract-auditor, blockchain-developer |
| data-ai | data-engineer, ml-engineer, data-scientist |
| database | postgres-specialist, redis-expert |
| devops-infrastructure | kubernetes-specialist, terraform-expert, cloud-architect |
| development-team | backend-architect, frontend-developer, mobile-developer |
| development-tools | code-optimizer, dependency-manager |
| game-development | unity-developer, godot-specialist |
| mcp-dev-team | mcp-server-architect, mcp-developer |
| modernization | legacy-modernizer, architecture-migrator |
| performance-testing | performance-engineer, web-vitals-optimizer |
| programming-languages | rust-expert, go-specialist, python-architect |
| security | security-auditor, penetration-tester |
| ui-analysis | screenshot-analyzer, accessibility-auditor |
| web-tools | seo-analyzer, accessibility-specialist |
Full list: 26 categories, 413+ agents
Refreshing the catalog
The category table in skills/template-catalog/SKILL.md
is auto-generated from the upstream cli-tool/components/agents tree (the same
source npx --agent installs from — not the incomplete docs/api/agents.json).
To resync counts and names after upstream changes:
node scripts/sync-catalog.mjs
# Optional: raise the GitHub API rate limit (60/h anon → 5000/h)
GITHUB_TOKEN=ghp_xxx node scripts/sync-catalog.mjs
Requires curl and Node 18+. The script rewrites only the block between the
<!-- catalog:start --> / <!-- catalog:end --> markers plus the agent/category
counts. It exits non-zero and leaves the file untouched if the upstream layout
changed or zero agents were found — no silently-wrong table.
Skill Priority
- Process — brainstorming, debugging, verification (always first)
- Implementation — TDD, code-review, frontend-design (during work)
- Templates — only when no existing skill covers the need
Anti-Patterns
- Skip brainstorming → "just code it"
- Write code before failing test
- Claim "fixed" without running verification
- Work without beads task
- Multiple fixes simultaneously
- Trust subagent reports without own verification
- Qualifier language ("should work", "probably fixed")
Installation
From GitHub (recommended)
claude plugin marketplace add maslennikov-ig/template-bridge
claude plugin install template-bridge@template-bridge-marketplace
From local clone
git clone https://github.com/maslennikov-ig/template-bridge.git
claude plugin marketplace add /path/to/template-bridge
claude plugin install template-bridge@template-bridge-marketplace
Three layers of protection
The plugin uses three layers to ensure Claude never forgets the workflow, even in a fresh context:
| Layer | Mechanism | How it works |
|---|---|---|
| 1. CLAUDE.md | System prompt (highest priority) | Rules loaded before anything else, every session |
| 2. SessionStart hook | Echo reminder after bd prime | Explicit reminder in context on session start |
| 3. Skill description | Plugin skill listing | unified-workflow trigger visible in skill list |
Layer 1: Copy CLAUDE.md to global config
The repo includes a ready-to-use CLAUDE.md. Copy its contents to your global config:
# Append to your global CLAUDE.md (create if doesn't exist)
cat /path/to/template-bridge/CLAUDE.md >> ~/.claude/CLAUDE.md
Or copy manually — the file is short (the full 9-step workflow + 4 rules).
Layer 2 (recommended): Hooks
The repo includes settings.example.json with two hooks:
- SessionStart —
bd prime(loads Beads context) + workflow reminder - PreCompact —
bd prime(re-injects Beads context before compaction so tasks survive)
Merge into your ~/.claude/settings.json:
# View the example first
cat /path/to/template-bridge/settings.example.json
# Then manually merge the "hooks" section into your ~/.claude/settings.json
Or copy the hooks block directly:
{
"hooks": {
"PreCompact": [
{
"matcher": "",
"hooks": [
{ "type": "command", "command": "bd prime" }
]
}
],
"SessionStart": [
{
"matcher": "",
"hooks": [
{ "type": "command", "command": "bd prime" },
{ "type": "command", "command": "echo 'WORKFLOW REMINDER: Invoke template-bridge:unified-workflow before any task. Flow: bd create → brainstorm → plan → TDD → review → verify → finish → bd close'" }
]
}
]
}
}
Layer 3: Automatic
The skill description is always visible in Claude's skill listing — no setup needed.
Updating
claude plugin install template-bridge@template-bridge-marketplace
Plugin re-reads from the source on reinstall.
Contributing
- Fork this repo
- Edit skills or add new ones
- Submit a PR
Plugin structure follows the official Claude Code plugin spec.
Credits
- Superpowers by Jesse Vincent — the TDD/brainstorming/debugging methodology
- Beads by Steve Yegge — task tracking for AI workflows
- claude-code-templates by Daniel Avila — 413+ specialist agent templates